Weekly Schedule

Weekly Schedule

plan for every week

Posted by Treaseven on December 30, 2024

12.30-1.5进度

论文阅读计划

  • Interstellar: Using Halide’s Scheduling Language to Analyze DNN Accelerators
  • Analytical Characterization and Design Space Exploration for Optimization of CNNs
  • Mind mappings: enabling efficient algorithm-accelerator mapping space search
  • Optimizing Deep Learning Inference via Global Analysis and Tensor Expressions
  • AStitch: Enabling a New Multi-dimensional Optimization Space for Memory-intensive ML Training and Inference on Modern SIMT Architectures
  • Hidet: Task-Mapping Programming Paradigm for Deep Learning Tensor Programs

论文复现工作

  • CMLCompiler
    决策树算法sklearn转换成tvm的DL model的流程如下：
    cmlcompiler.model → build_model → init_model → init → parse_params → convert_decision_tree → __get_algo → decision_tree_classifier →

  • Mind Mappings

本周进度还是相对较慢，因为中间元旦放假两天，耽误了一些进度，下周还是需要更加加油！！！

1.6-1.12进度

论文阅读计划

  • ROLLER: Fast and Efficient Tensor Compilation for Deep Learning
  • SmartMem: Layout Transformation Elimination and Adaptation for Efficient DNN Execution on Mobile
  • MonoNN: Enabling a New Monolithic Optimization Space for Neural Network Inference Tasks on Modern GPU-Centric Architectures
  • Accelerated Auto-Tuning of GPU Kernels for Tensor Computations
  • A full-stack search technique for domain optimized deep learning accelerators
  • RAMMER: Enabling Holistic Deep Learning Compiler Optimizations with rTasks
  • Bridging the Gap Between Domain-specific Frameworks and Multiple Hardware Devices

论文复现工作

  • CMLCompiler
    测试代码编写：测试的时候在用测试集，只需利用测试集中的一行数据进行测试，关注的是单次推理的性能，减少内存占用，之前的想法是用整个测试集进行测试 CMLCompiler的代码全部阅读完，但目前发现CMLCompiler在复现的时候效果不如论文中的结果，现在考虑重新写CMLCompiler的代码

  • Heron
    CPU的实验 config属性: out_name、method、max_trials、runner_number、runner_repeat、runner_timeout、build_timeout、in_dtype、out_dtype、cases、target_name、codegen_type、get_op、device_id、tuned

这周状态

  • 这周前四天效率还是比较高，但是后面三天效率比较低，下周还是要坚持6天效率比较高，加油

每周学习的计划

  • 这周没有制定相应的学习计划，但实际应该制定一些相应的学习计划，比如TVM源码的学习笔记，从下周开始必须整理相应的TVM源码的学习计划内容

1.13-1.19进度

论文阅读计划

  • DOPpler: Parallel Measurement Infrastructure for Auto-Tuning Deep Learning Tensor Programs
  • Mind the Gap: Attainable Data Movement and Operational Intensity Bounds for Tensor Algorithms
  • TIRAMISU: A Polyhedral Compiler for Expressing Fast and Portable Code
  • DISTAL: The Distributed Tensor Algebra Compiler
  • AKG: Automatic Kernel Generation for Neural Processing Units using Polyhedral Transformations
  • Tlp: A deep learning-based cost model for tensor program tuning
  • DNNFusion: Accelerating Deep Neural Networks Execution with Advanced Operator Fusion

1.20-1.22进度

论文阅读计划

  • Chimera: An Analytical Optimizing Framework for Effective Compute-intensive Operators Fusion
  • UNIT: Unifying tensorized instruction compilation
  • FreeTensor: A Free-Form DSL with Holistic Optimizations for Irregular Tensor Programs
  • Uncovering Nested Data Parallelism and Data Reuse in DNN Computation with FractalTensor
  • MCFuser: High-Performance and Rapid Fusion of Memory-Bound Compute-Intensive Operators
  • Effectively Scheduling Computational Graphs of Deep Neural Networks toward Their Domain-Specific Accelerators
  • Apollo: Automatic Partition-based Operator Fusion through Layer by Layer Optimization

2025年每周工作进度表

-2.16进度

论文阅读计划

  • A Comparison of End-to-End Decision Forest Inference Pipelines
  • SilvanForge A Schedule Guided Retargetable Compiler for Decision Tree Inference
  • Treebeard An Optimizing Compiler for Decision Tree Based ML Inference
  • Accelerating Decision-Tree-based Inference through Adaptive Parallelization
  • Tahoe Tree Structure-Aware High Performance Inference Engine for Decision Tree Ensemble on GPU

论文复现工作

  • Heron: 目前将该代码中的CPU部分的实验过程全部梳理了一遍，但对调度规则制定和调度搜索这一块的代码还没有具体详细看，接下来继续整理这一块的内容，然后在比较GPU部分的实验
  • Hummingbird: 代码中关于tvm部分的实验梳理一遍，还没有准确定位关于决策树三种算法的选择块的代码，需要跑通这一块GPU、CPU部分的实验，作为后续对比实验的数据,目前发现我在做实验过程利用hummingbird这个代码对决策树进行推理的效果不如sklearn好，使用hummingbird的版本为0.4.12，tvm的版本为0.15.0，然后运行hummingbird在tvm上会报下面的错误，估计这个错跟tvm的版本有关，在构建计算图的时候发生了一些错误导致的，可以考虑tvm版本的事情，选择一个稳定的tvm版本来做实验，另外论文中使用的tvm版本是0.6.0，这里建议和github仓库维护者反应一下这个问题 
    [15:51:03] /home/newuser/tvm/src/te/schedule/bound.cc:119: not in feed graph consumer = compute(p0_red_temp, body=[T.reduce(T.comm_reducer(lambda argmax_lhs_0, argmax_lhs_1, argmax_rhs_0, argmax_rhs_1: (T.Select(argmax_lhs_1 > argmax_rhs_1 or argmax_lhs_1 == argmax_rhs_1 and argmax_lhs_0 < argmax_rhs_0, argmax_lhs_0, argmax_rhs_0), T.Select(argmax_lhs_1 > argmax_rhs_1, argmax_lhs_1, argmax_rhs_1)), [-1, T.float32(-3.4028234663852886e+38)]), source=[k1, p0[ax0, k1]], init=[], axis=[T.iter_var(k1, T.Range(0, 251), "CommReduce", "")], condition=T.bool(True), value_index=0), T.reduce(T.comm_reducer(lambda argmax_lhs_0, argmax_lhs_1, argmax_rhs_0, argmax_rhs_1: (T.Select(argmax_lhs_1 > argmax_rhs_1 or argmax_lhs_1 == argmax_rhs_1 and argmax_lhs_0 < argmax_rhs_0, argmax_lhs_0, argmax_rhs_0), T.Select(argmax_lhs_1 > argmax_rhs_1, argmax_lhs_1, argmax_rhs_1)), [-1, T.float32(-3.4028234663852886e+38)]), source=[k1, p0[ax0, k1]], init=[], axis=[T.iter_var(k1, T.Range(0, 251), "CommReduce", "")], condition=T.bool(True), value_index=1)], axis=[T.iter_var(ax0, T.Range(0, 103069), "DataPar", "")], reduce_axis=[T.iter_var(k1, T.Range(0, 251), "CommReduce", "")], tag=comm_reduce_idx, attrs={})
  • CMLCompiler: 1. 发现中关于决策树转换部分的代码还存在问题，同时需要决策树变换结果打印出来，仔细分析一下是否还可以采用调度方法来提高，目前打算这个代码作为一个baseline然后进行修改; 2.决策树训练的问题，之前是使用yearPredicition数据集里面的全部特征，后面和作者联系发现作者在训练决策树模型只使用了27个特征，同时也约束了决策树的深度，解决一开始直接用决策树去训练得到的一个模型然后在用cmlcompiler会出现内存溢出的问题；3.关于上面hummingbird出现的错在cmlcompiler也会出现，然后发现确实是在构建计算图的时候出错，然后我对tree_common.py文件中tree_gemm这个函数进行修改，出错的原因是relay.argmax的规约操作导致在TVM的调度时出现边界推断问题，从而导致编译失败，修改的操作将argmax分解为更基本的操作，明确指定了所有中间步骤的形状和类型

** TVM源码学习 **

目前在看TVM官网上的开发教程，这周还没有怎么看，下周我一定要好好看

2.17-2.23进度

论文阅读计划

  • One-Shot Tuner for Deep Learning Compilers
  • A Flexible Approach to Autotuning Multi-Pass Machine Learning Compilers
  • Transfer-Tuning: Reusing Auto-Schedules for Efficient Tensor Program Code Generation
  • Effective Performance Modeling and Domain-Specific Compiler Optimization of CNNs for GPUs
  • Alcop: Automatic load-compute pipelining in deep learning compiler for ai-gpus
  • Automatic Generation of Multi-Objective Polyhedral Compiler Transformations
  • Welder: Scheduling Deep Learning Memory Access via Tile-graph
  • Bolt: Bridging the gap between auto-tuners and hardware-native performance
  • A Holistic Functionalization Approach to Optimizing Imperative Tensor Programs in Deep Learning
  • Chimera: An Analytical Optimizing Framework for Effective Compute-intensive Operators Fusion
  • Apollo: Automatic Partition-based Operator Fusion through Layer by Layer Optimization
  • Fasor: A Fast Tensor Program Optimization Framework for Efficient DNN Deployment

论文复现工作

  • One-Shot Tuner: 这个还没有看，但是打算用这个代价模型融合到Heron里面
  • Heron: 目前在跑CPU部分的实验，本次重点了解代码中关于规则制定部分的内容，但是现在跑的实验还存在问题，就是我自己的电脑CPU和代码要求的不一致，我的电脑是不能用avx5只能用avx2,所以导致平台设置需要改变一下，另外代码要求的llvmshi 0.8版本，我的版本也不一致需要改一下

2.24-3.2进度

论文阅读计划

  • A Learned Performance Model For Tensor Processing Units
  • DNNFusion: Accelerating Deep Neural Networks Execution with Advanced Operator Fusion
  • Optimal Kernel Orchestration for Tensor Programs with Korch
  • Chimera: An Analytical Optimizing Framework for Effective Compute-intensive Operators Fusion
  • UNIT: Unifying tensorized instruction compilation
  • Lorien: Efficient Deep Learning Workloads Delivery
  • A Practical Tile Size Selection Model for Affine Loop Nests
  • AKG: Automatic Kernel Generation for Neural Processing Units using Polyhedral Transformations
  • Bring Your Own Codegen to Deep Learning Compiler
  • A Deep Learning Based Cost Model for Automatic Code Optimization

论文复现工作

  • Heron: 重新梳理一下context类里面值的赋值问题，之前代码中存在如dense_i与P#ST:dense,AX:i互相乱用的问题，从而导致在生成schedule.py文件的时候发生字符串错乱的问题，目前该部分的代码都已经修改完整，目前是在CPU部分的实验，考虑GPU部分的实验也存在相应的问题在看GPU部分的代码需提前考虑一下；另外也了解和熟悉目前tvm存在的调度方法以及在该代码中采用的部分，目前在看部分的代码还是属于论文中的Generation部分，接下来开始看Exploration部分的代码，另外目前代码由于实验平台的限制导致不能采用tensorize调度，因为目前电脑的CPU支持avx2不支持avx512,tvm源码中关于tensorize部分的代码没有我电脑合适的张量化的方法，因此我在代码中是禁用tensorize，另外我生成的constraints.py文件中dense_global_tileAll = model.NewIntVar(1, 64, ‘dense_global_tileAll’)，与作者在网上提供的constraints.py有点不一致，考虑是这个原因导致后面会发生Illegal instruction (core dumped)的问题

  • CMLCompiler: 重新分析Hummingbird中关于决策树转化为矩阵的过程，分析目前CMLCompiler中没有关注到的点，关于Hummingbird中关于决策树转化部分的代码目前还没有看(下周可以具体分析)，目前在分析后面步骤的优化，另外在其实对这个想法的优化其实是集中对于GEMM算子的优化，另外他这个转换其实也是一个小的神经网络模型，我是不是应该考虑这个步骤分别对应于计算密集型算子、内存密集型算子，另外现在的想法没有结合决策树本身的特性，以及分析一下树的结构的工作下周继续开展一下，整理一下初步的小实验

TVM源码学习 整理学习TVM代码库实例讲解部分，目前初步对TVM整个代码结构有一定的了解，重点学习TVM中关于C++与Python之间如何相互调用的过程，学习PackedFunc的内容，关于下周可以结合Felix项目来配合TVM的源码学习

这周状态 总体而言还是比较不错，不过最近稍微有点焦虑，在思考自己的课题进度以及实验idea遇到一些困难，有点太敏感外界因素的影响，其实自己应该放下包袱，你没有什么好担心，专专心心地学习才是最重要的不要被外界的事情所干扰，沉下心来学习才是最重要的，另外不要太关注别人的进度，把握好自己的进度，每天的自己比昨天的自己有进步就是很棒的，不要拿自己去任何比较，这样既不会给自己带来任何的进步反而会影响自己的学习状态，赵凯加油，坚持下来，胜利是属于能坚持到最后的人的！！！好好规划好自己的每一天，让自己每一天都过得充实，而不是忙忙碌碌的一天

学习计划 每周每天还是要给自己制定相应的学习计划，坚持坚持坚持！！！

3.3-3.9进度

论文阅读计划

  • DISTAL: The Distributed Tensor Algebra Compiler
  • Optimizing the Memory Hierarchy by Compositing Automatic Transformations on Computations and Data
  • Modeling the Interplay between Loop Tiling and Fusion in Optimizing Compilers Using Affine Relations
  • MCFuser: High-Performance and Rapid Fusion of Memory-Bound Compute-Intensive Operators
  • Fireiron: A Data-Movement-Aware Scheduling Language for GPUs
  • DREW: Efficient Winograd CNN Inference with Deep Reuse
  • DeepCuts: A Deep Learning Optimization Framework for Versatile GPU Workloads
  • HASCO: Towards Agile HArdware and Software CO-design for Tensor Computation
  • FusionStitching: Boosting Memory Intensive Computations for Deep Learning Workloads
  • Atomic Dataflow based Graph-Level Workload Orchestration for Scalable DNN Accelerators
  • Nimble: Lightweight and Parallel GPU Task Scheduling for Deep Learning
  • Tlp: A deep learning-based cost model for tensor program tuning

论文复现工作

  • Heron： 目前是跑通Heron/tests/quick_start/dlboost中的实验，验证之前关于Context、KnobManager、Solver类中的变量赋值问题，之前作者代码存在字符串混用的问题，

  • Hummingbird dataset iris datadir benchmark/operators/datasets/ modeldir benchmarks/operators/models/ operator all backend all(onnx-ml、hb-torchscript、hb-tvm) cpus 1 batch_size 1000000 gpu false output none nrows 1000000 niters 5 validate false extra {}

3.10-3.16进度

论文阅读计划

  • GTuner: Tuning DNN Computations on GPU via Graph Attention Network
  • Effectively Scheduling Computational Graphs of Deep Neural Networks toward Their Domain-Specific Accelerators
  • Optimizing DNN Computation with Relaxed Graph Substitutions
  • AutoGraph: Optimizing DNN Computation Graph for Parallel GPU Kernel Execution
  • POET: Training Neural Networks on Tiny Devices with Integrated Rematerialization and Paging
  • Collage: Seamless Integration of Deep Learning Backends with Automatic Placement
  • Exploiting Subgraph Similarities for Efficient Auto-tuning of Tensor Programs
  • Transferable Graph Optimizers for ML Compilers
  • Optimizing DNN computation graph using graph substitutions
  • IOS: Inter-Operator Scheduler for CNN Acceleration
  • Equality Saturation for Tensor Graph Superoptimization
  • Pruner: A Speculative Exploration Mechanism to Accelerate Tensor Program Tuning

论文复现工作

  • Heron: 这周搞定Heron的GPU部分实验，目前已经跑通实验效果与作者提供的一致，发现作者代码中存在的字符串乱用bug目前已经完成修改，同时针对之前没发现CPU部分实验问题进行修改，之前存在的问题是轴区间访问问题，之前不能正确获取相应的值，对其进行修改将CPU部分实验效果进行改善，另外研究Heron中利用调度原语的作用，分析各个调度原语所带来的优化效果，对作者的调度模板生成算法进行进一步了解

学习计划 下周开始每晚学习一下算法进阶课，需要补一下算法方面的知识，同时下周开始每天开始写日报

3.17-3.23进度

论文阅读计划

  • IMTP: Search-based Code Generation for In-memory Tensor Programs
  • Gensor: A Graph-based Construction Tensor Compilation Method for Deep Learning
  • Optimizing Dynamic-Shape Neural Networks on Accelerators via On-the-Fly Micro-Kernel Polymerization
  • Hector: An Efficient Programming and Compilation Framework for Implementing Relational Graph Neural Networks in GPU Architectures
  • SoD2: Statically Optimizing Dynamic Deep Neural Network Execution
  • TensorMap: A Deep RL-Based Tensor Mapping Framework for Spatial Accelerators
  • Sifter An Efficient Operator Auto-Tuner with Speculative Design Space Exploration for Deep Learning Compiler

论文复现工作

  • Heron: 搞清楚Heron的代码生成规则，对TVM的代码生成模板目前已经掌握，同时也了解GPU的架构，对于生成规则的GPU的约束
  • AMOS: 目前成功安装，对AMOS的代码架构了解已经完成
  • FlexTensor: 目前还没有看，只运行成功测试脚本，保证环境已成功完成 本周还了解TVM的python与C++端的代码如何配合完成

算法学习 完成搜索算法的模拟退火、爬山法、bfs

4.21-4.27进度

论文阅读计划

  • MapZero: Mapping for Coarse-grained Reconfigurable Architectures with Refinforcement Learning and Monte-Carlo Tree Search
  • WACO: Learning Workload-Aware Co-optimization of the Format and Schedule of a Sparse Tensor Program
  • SparseTIR: Composable Abstractions for Sparse Compilation in Deep Learning
  • vMCU: Coordinated Memory Management and Kernel Optimization for DNN Inference on MCUs
  • Automatic Generation of Vectorizing Compilers for Customizable Digital Signal Processors
  • Graphene: An IR for Optimized Tensor Computations on GPUs
  • Hydride: A Retargetable and Extensible Synthesis-based Compiler for Modern Hardware Architectures
  • EVT: Accelerating Deep Learning Training with Epilogue Visitor Tree

Bayesian Optimization (ICLR-2025) Latent Bayesian Optimization via Autoregressive Normalizing Flows Standard Gaussian Process is All You Need for High-Dimensional Bayesian Optimization Second-Order Min-Max Optimization with Lazy Hessians AFlow: Automating Agentic Workflow Generation Test-time Alignment of Diffusion Models without Reward Over-optimization PABBO: Preferential Amortized Black-Box Optimization Bayesian Optimization of Antibodies Informed by a Generative Model of Evolving Sequences Nesterov acceleration in benignly non-convex landscapes Mitigating Information Loss in Tree-Based Reinforcement Learning via Direct Optimization Bayesian Experimental Design Via Contrastive Diffusions Adaptive backtracking for faster optimization Searching for Optimal Solutions with LLMs via Bayesian Optimization Pareto Prompt Optimization Complexity Lower Bounds of Adaptive Gradient Algorithms for Non-convex Stochastic Optimization under Relaxed Smoothness Sharpness-Aware Black-Box Optimization Optimizing Posterior Samples for Bayesian Optimization via Rootfinding Causal Discovery via Bayesian Optimization Selective Task Group Updates for Multi-Task Optimization Few for Many: Tchebycheff Set Scalarization for Many-Objective Optimization ParetoFlow: Guided Flows in Multi-Objective Optimization BOFormer: Learning to Solve Multi-Objective Bayesian Optimization via Non-Markovian RL Multi-objective antibody design with constrained preference optimization

ICML-2024 Principled Preferential Bayesian Optimization High-Dimensional Bayesian Optimization via Semi-Supervised Learning with Optimized Unlabeled Data Sampling MALIBO: Meta-learning for Likelihood-free Bayesian Optimization In-Context Freeze-Thaw Bayesian Optimization for Hyperparameter Optimization BRAIn: Bayesian Reward-conditioned Amortized Inference for natural language generation from feedback BOtied: Multi-objective Bayesian optimization with tied multivariate ranks Bayesian Knowledge Distillation: A Bayesian Perspective of Distillation with Uncertainty Quantification Random Exploration in Bayesian Optimization: Order-Optimal Regret and Computational Efficiency Is In-Context Learning in Large Language Models Bayesian? A Martingale Perspective Probability Distribution of Hypervolume Improvement in Bi-objective Bayesian Optimization A Unified View of FANOVA: A Comprehensive Bayesian Framework for Component Selection and Estimation Bayesian Exploration Networks Bayesian Uncertainty for Gradient Aggregation in Multi-Task Learning A Bayesian Approach to Online Planning Deep Functional Factor Models: Forecasting High-Dimensional Functional Time Series via Bayesian Nonparametric Factorization Accelerating Look-ahead in Bayesian Optimization: Multilevel Monte Carlo is All you Need More Flexible PAC-Bayesian Meta-Learning by Learning Learning Algorithms Bayesian Design Principles for Offline-to-Online Reinforcement Learning Accelerating Convergence in Bayesian Few-Shot Classification PAC-Bayesian Generalization Bounds for Knowledge Graph Representation Learning Boundary Exploration for Bayesian Optimization With Unknown Physical Constraints Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior A Sober Look at LLMs for Material Discovery: Are They Actually Good for Bayesian Optimization Over Molecules Partially Stochastic Infinitely Deep Bayesian Neural Networks Vanilla Bayesian Optimization Performs Great in High Dimensions

NeurIPS-2024 Sample-efficient Bayesian Optimisation Using Known Invariances Improved Bayes Regret Bounds for Multi-Task Hierarchical Bayesian Bandit Algorithms Online Bayesian Persuasion Without a Clue Bayesian Optimisation with Unknown Hyperparameters: Regret Bounds Logarithmically Closer to Optimal Bayesian Domain Adaptation with Gaussian Mixture Domain-Indexing Transition Constrained Bayesian Optimization via Markov Decision Processes Stopping Bayesian Optimization with Probabilistic Regret Bounds Principled Bayesian Optimization in Collaboration with Human Experts Amortized Bayesian Experimental Design for Decision-Making Bayesian Strategic Classification Cost-aware Bayesian Optimization via the Pandora’s Box Gittins Index Conjugate Bayesian Two-step Change Point Detection for Hawkes Process General bounds on the quality of Bayesian coresets Minimizing UCB: a Better Local Search Strategy in Local Bayesian Optimization A survey and benchmark of high-dimensional Bayesian optimization of discrete sequences

CATALOG
    FEATURED TAGS
    Tensor Compiler Compiler Optimization Code Generation Heterogeneous Systems Operator Fusion Deep Neural Network Recursive Tensor Execution Deep Learning Compiler Classical Machine Learning Compiler Optimizations Bayesian Optimization Autotuning Spatial Accelerators Tensor Computations Code Reproduction Neural Processing Units Polyhedral Model Auto-tuning Machine Learning Compiler Neural Network Program Transformations Tensor Programs Deep learning Tensor Program Optimizer Search Algorithm Compiler Infrastructure Scalalbe and Modular Compiler Systems Tensor Computation GPU Task Scheduling GPU Streams Tensor Expression Language Automated Program optimization Framework AI compiler memory hierarchy data locality tiling fusion polyhedral model scheduling domain-specific architectures memory intensive TVM Sparse Tensor Algebra Sparse Iteration Spaces Optimizing Transformations Tensor Operations Machine Learning Model Scoring AI Compiler Memory-Intensive Computation Fusion Neural Networks Dataflow Domain specific Language Programmable Domain-specific Acclerators Mapping Space Search Gradient-based Search Deep Learning Systems Systems for Machine Learning Programming Models Compilation Design Space Exploration Tile Size Optimization Performance Modeling High-Performance Tensor Program Tensor Language Model Tensor Expression GPU Loop Transformations Vectorization and Parallelization Hierarchical Classifier TVM API Optimizing Compilers Halide Pytorch Optimizing Tensor Programs Gradient Descent debug Automatic Tensor Program Tuning Operators Fusion Tensor Program Cost Model Weekly Schedule Spatio-temporal Schedule tensor compilers auto-tuning tensor program optimization compute schedules Tensor Compilers Data Processing Pipeline Mobile Devices Layout Transformations Transformer Design space exploration GPU kernel optimization Compilers Group Tuning Technique Tensor Processing Unit Hardware-software Codeisgn Data Analysis Adaptive Systems Program Auto-tuning python api Code Optimization Distributed Systems High Performance Computing code generation compiler optimization tensor computation Instructions Integration Code rewriting Tensor Computing DSL CodeReproduction Deep Learning Compiler Loop Program Analysis Nested Data Parallelism Compute-Intensive Automatic Exploration Loop Fusion Data Movement C++ Machine Learning System Decision Forest Optimizfing Compiler Decision Tree Ensemble Decision Tree Inference Parallelization Optimizing Compiler decision trees random forest machine learning parallel processing multithreading Tree Structure Performance Model Code generation Compiler optimization Tensor computation accelerator neural networks optimizing compilers autotuning performance models deep neural networks compilers auto-scheduling tensor programs Tile size optimization Performance modeling Program Functionalization affine transformations loop optimization Performance Optimization Subgraph Similarity deep learning compiler Intra- and Inter-Operator Parallelisms ILP tile-size operator fusion cost model graph partition zero-shot tuning tensor program kernel orchestration machine learning compiler Loop tiling Locality Polyhedral compilation Optimizing Transformation Sparse Tensors Asymptotic Analysis Automatic Scheduling Optimization Operation Fusion data reuse deep reuse Tensorize docker graph substitution compiler Just-in-time compiler graph Tensor program construction tensor compilation graph traversal Markov analysis Deep Learning Compilation Tensor Program Auto-Tuning Decision Tree Search-based code generation Domain specific lanuages Parallel architectures Dynamic neural network mobile device spatial accelerate software mapping reinforcement learning Computation Graph Graph Scheduling and Transformation Graph-level Optimization Operator-level Optimization Partitioning Algorithms IR Design Parallel programming languages Software performance Digitial signal processing Retargetable compilers Equational logic and rewriting Tensor-level Memory Management Code Generation and Optimizations Scheduling Sparse Tensor Auto-Scheduling Tensor Coarse-Grained Reconfigurable Architecture Graph Neural Network Reinforcement Learning Auto-Tuning Domain-Specific Accelerator Deep learning compiler Long context Memory optimization code analysis